Improving Traffic Operations Under Incidents Using Large-Scale Traffic Simulations And Simulation-Based Optimization Methods
dc.contributor.advisor | Zhu, Shanjiang | |
dc.contributor.author | Liu, Guanqi | |
dc.creator | Liu, Guanqi | |
dc.date | 2019-05-03 | |
dc.date.accessioned | 2019-05-31T14:40:32Z | |
dc.date.available | 2021-05-03T06:48:57Z | |
dc.description | This dissertation has been embargoed for 2 years. It will not be available until May 2021 at the earliest. | |
dc.description.abstract | Traffic incidents are a major contributing factor towards traffic congestion. Effective incident response strategies can significantly reduce non-recurrent congestion and thus represent important research topics. In many cases, traffic operators only consider incident response strategies for local conditions on a relative basis. There is a need for efficient algorithms that could help identify optimal traffic operation control strategies on a regional network with complex traffic dynamics fully considered. However, identifying such optimal incident response strategies is a complex problem because of the large number of possible scenarios due to the combination of different type of control strategies, multiple time periods, different control parameters, multiple locations, and traffic dynamics. This optimization problem is high-dimensional and nonlinear. The underlying traffic dynamics do not have a close-form objective function and are also very time-consuming to evaluate. To address these challenges, an integrated optimization framework that combines traffic simulation model with Simulation-Based Optimization (SBO) methods will be developed to identify the optimal traffic operation strategies under incident conditions. SBO methods can approximate the input-output relationship of a complex system using information from a set of sampling points and improve the response surface (objective function) effectively using new inputs guided from previous knowledge. Surrogate Models, a sub-category of SBO family, is a powerful method for solving optimization problems where the response surface is complex and unknown. The proposed integrated optimization framework was tested by combining Surrogate Models with a small analytical assignment network where the theoretical optimal can be analytically derived. Three Surrogate Models were evaluated: Quadratic Polynomial Regression (QPR), Radial Basis Function (RBF) and Bayesian Optimization (BO).The performance of the proposed framework is discussed. Results showed that BO has the best optimal solution searching capability yet with great running time. Though QPR requires much less computational time, it was shown to have poor optimal solution searching capability. This research then integrated a well calibrated traffic simulation model with SBO methods to identify the optimal traffic operation strategies under incident conditions on an urban area roadway network in Northern Virginia and the network-wise performance of the proposed framework was studied. This study presented a sequential method to further reduce the research space and improve the efficiency of the proposed model. Three different SBO algorithms are tested and compared, the performance of the proposed framework is discussed. The optimization results show the significant improvements of system total travel time and corridor congestion pattern. The results indicated that the proposed optimization model is capable of enhancing response strategies that are high dimensional and with an unknown objective function, solving both of which by either derivative optimization algorithms or brute-force searching method is infeasible. This study then introduced multiple choices of traffic operation strategies (Ramp Metering, Shoulder Lane, and Variable Message Sign) that function cooperatively and simultaneously to respond incident related traffic congestion. The optimal traffic operation strategy was then identified by using the proposed optimization framework. The results also showed significant improvements in total travel time in the system when multiple traffic operation strategies are implemented under incident scenarios. Concurrent deployment of synergistic multiple operational strategies are found to outperform each of the strategies working in isolation. This research is one of the first studies that aims to identify the optimal incident response strategies with multiple choices of traffic operation controls for responding to non-recurrent congestion on a large scale time-dependent simulation network with detailed traffic dynamics (e.g. users’ reactions, queuing propagation, bottlenecks) fully considered. | |
dc.identifier.uri | https://hdl.handle.net/1920/11432 | |
dc.subject | Incident Response Management | |
dc.subject | Simulation-Based Optimization | |
dc.subject | Surrogate Model | |
dc.subject | Traffic Operation | |
dc.subject | Traffic Simulation Model | |
dc.title | Improving Traffic Operations Under Incidents Using Large-Scale Traffic Simulations And Simulation-Based Optimization Methods | |
dc.type | Dissertation | |
thesis.degree.discipline | Civil and Infrastructure Engineering | |
thesis.degree.grantor | George Mason University | |
thesis.degree.level | Doctoral | |
thesis.degree.name | Doctor of Philosophy in Civil and Infrastructure Engineering |